runner

run_experiment - orchestrates a single ExperimentConfig end-to-end.

Sequence per RunConfig:

Build data store + log store from ExperimentConfig top-level specs.
Read raw rows from data.source.
Apply transforms in order.
Build backend; backend.prepare(model=..., run_dir=...).
Build algorithm from registry + run.algorithm.params; instantiate.
algorithm.train(RunContext) -> RunResult.
backend.teardown(handles).
Run eval if enabled. (Eval is best-effort; a failure here doesn't kill the run.)
Persist run_result.json.

Returns a list[RunResult] (one per run in the experiment).

attributelogger

= logging.getLogger(__name__)

func_build_from_spec(getter, spec) -> Any

paramgetter

paramspec

Returns

typing.Any

func_load_rows(data, data_store) -> list[dict[str, Any]]

paramdataDataConfig

paramdata_store

Returns

list[dict[str, typing.Any]]

func_apply_transforms(rows, data) -> list[dict[str, Any]]

paramrowslist[dict[str, Any]]

paramdataDataConfig

Returns

list[dict[str, typing.Any]]

func_persist_result(run_dir, result, hparams) -> None

paramrun_dirPath

paramresultRunResult

paramhparamsdict[str, Any]

Returns

None

func_execute_run(*, cfg, run, base_output_dir, extra_context=None) -> RunResult

paramcfgExperimentConfig

paramrunRunConfig

parambase_output_dirPath

paramextra_contextdict[str, Any] | None

= None

Returns

evsys_sdk.protocols.RunResult

funcrun_experiment(cfg_or_path, *, extra_context=None) -> list[RunResult]

Run an experiment from a parsed config or YAML file.

extra_context is merged into each run's RunContext.extras - used by Experiment to pass the dashboard store + dashboard_run_id so in-loop validation can upload its rollouts.

paramcfg_or_path

paramextra_contextdict[str, Any] | None

= None

Returns

list[evsys_sdk.protocols.RunResult]